AI Paper Atlas

Research map / Image generation / Text-to-image diffusion

StableRep: Synthetic Images from Text-to-Image Models Make Strong Visual Representation Learners

Yonglong Tian, Lijie Fan, Phillip Isola, Huiwen Chang, Dilip Krishnan

NeurIPS 2023 · Image generation › Text-to-image diffusion · approach: Inference-time T2I improvement

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Train StableRep with synthetic images and multi-positive same-caption contrastStableRep vs CLIP trained on real images · ImageNet linear probe, CC12Maccuracy: 70.3 -> 73.5Improved
Train StableRep with synthetic images and multi-positive same-caption contrastStableRep vs CLIP trained on real images · ImageNet linear probe, RedCapsaccuracy: 71.9 -> 74.5Improved
Replace real images with synthetic images for SimCLR pretrainingSimCLR · ImageNet linear probe, CC12Maccuracy: 61.5 -> 63.7Improved
Replace real images with synthetic images for CLIP pretrainingCLIP · ImageNet linear probe, CC12Maccuracy: 70.3 -> 67.8Hurt
Generate multiple images per caption instead of one (ablation)StableRep · ImageNet linear probe, CC3M, fixed image-generation budgetaccuracy: 61.2 -> 66.2Improved
Increase images per caption in a batch from 1 to 8 (ablation)StableRep · ImageNet linear probe, CC3M, fixed batch sizeaccuracy: 60.5 -> 69.8Improved
Use smaller generation guidance scales w in {2,3} instead of w in {8,10} (ablation)StableRep · ImageNet linear probe, CC12Maccuracy: 70.8 -> 72.8Improved
Scale the backbone from ViT-B/16 to ViT-L/16 (ablation)StableRep · ImageNet linear probe, CC12Maccuracy: 72.8 -> 74.7Improved
Train StableRep longer, from 35 to 105 epochs (ablation)StableRep · ImageNet linear probe, CC12Maccuracy: 72.8 -> 75.7Improved
Add random downsample augmentation (ablation)StableRep · Fine-grained linear probing average, CC12Maccuracy: 82.2 -> 83.7Improved
Add a non-affine BatchNorm layer to the classifier (ablation)StableRep trained for 35 epochs · ImageNet linear probeaccuracy: 71.4 -> 73.5Improved
Add a non-affine BatchNorm layer to the classifier (ablation)StableRep trained for 105 epochs · ImageNet linear probeaccuracy: 75.4 -> 75.2No clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy